Purpose: In team sports, tracking technologies are widely utilized to collect the positional features of the athletes thus quantifying physical load demands and the tactical behavior characteristics of interest. Tracking procedures are mostly based on global positioning systems (GPS), multiple camera setups and radio-frequency based systems (Coutts et al.. 2010: Di Salvo et al.. 2006: Frencken et al., 2010). In the last years, the employment of computer artificial intelligence, relying on deep learning algorithmic procedures has opened up new possible solution for sport movement tracking. In this context, very recently a new fully automatic tracking system (Track160`) combining computer-vision-based and iso-inertial measurement units inputs has been developed for collecting and providing accurate positional information. In view of the evident interest about accurate tracking solutions (Memmert et al.. 2016). an important first step is to investigate and establish the reliability and validity of the developed system. Therefore, the main purpose of this study was to investigate the reliability and validity of the Track160* system. Methods: A repeated measures design was used for investigating the reliability of the Track160` system outcomes and the agreement with those of a ground truth reference. Four football players performed a variety of locomotor activities like walking, jogging, fast running, sprinting along linear and multidirectional paths both as individuals and as couples with the between-subject distance kept known and constant. Moreover, the same trials and conditions where performed within a real game scenario of 5 vs 5 small sided games. The collected variables were: individual distances covered during different activities and through different paths, and between-subject positional relationships both in free view and occluded conditions. Results: Intra-subject, inter-subject and test-retest reliability of the distance measures were very high as shown with intra-class correlations (ICC) scores ranging between 0.97 and 0.99. The percentage difference between the ground truth reference and the outcomes of the tracking system was 1.2±0.35% and 1.4±0.42% for the free view and occluded conditions, respectively. The percentage difference between the known between-subject distance reference and the tracking system outcomes was 1.3±1.4% and 1.5±1.8% for the free view and occluded conditions, respectively. Conclusion: In light of these results, the Track160* system could be considered acceptable for capturing players`displacements in outdoor team sports such as football. The system showed high reliability and agreement scores when compared to ground truth references. Furthermore, the mean absolute error did not change with increasing movement speeds and occluded view conditions. The Track160* system provides accurate and reliable positional details and offer likely opportunities to develop new performance indicators from an individual and collective basis.
© Copyright 2018 World Congress of Performance Analysis of Sport XII. Veröffentlicht von Faculty of Kinesiology, University of Zagreb, Croatia. Alle Rechte vorbehalten.